quick_entrypoints_arm64.S revision 4359e61927866c254bc2d701e3ea4c48de10b79c
1/* 2 * Copyright (C) 2014 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17#include "asm_support_arm64.S" 18 19#include "arch/quick_alloc_entrypoints.S" 20 21 22 /* 23 * Macro that sets up the callee save frame to conform with 24 * Runtime::CreateCalleeSaveMethod(kSaveAll) 25 */ 26.macro SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 27 adrp xIP0, :got:_ZN3art7Runtime9instance_E 28 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 29 30 // Our registers aren't intermixed - just spill in order. 31 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 32 33 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 34 // Loads appropriate callee-save-method. 35 ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVE_FRAME_OFFSET ] 36 37 sub sp, sp, #176 38 .cfi_adjust_cfa_offset 176 39 40 // Ugly compile-time check, but we only have the preprocessor. 41#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVE != 176) 42#error "SAVE_ALL_CALLEE_SAVE_FRAME(ARM64) size not as expected." 43#endif 44 45 // Stack alignment filler [sp, #8]. 46 // FP callee-saves. 47 stp d8, d9, [sp, #16] 48 stp d10, d11, [sp, #32] 49 stp d12, d13, [sp, #48] 50 stp d14, d15, [sp, #64] 51 52 // GP callee-saves 53 stp x19, x20, [sp, #80] 54 .cfi_rel_offset x19, 80 55 .cfi_rel_offset x20, 88 56 57 stp x21, x22, [sp, #96] 58 .cfi_rel_offset x21, 96 59 .cfi_rel_offset x22, 104 60 61 stp x23, x24, [sp, #112] 62 .cfi_rel_offset x23, 112 63 .cfi_rel_offset x24, 120 64 65 stp x25, x26, [sp, #128] 66 .cfi_rel_offset x25, 128 67 .cfi_rel_offset x26, 136 68 69 stp x27, x28, [sp, #144] 70 .cfi_rel_offset x27, 144 71 .cfi_rel_offset x28, 152 72 73 stp x29, xLR, [sp, #160] 74 .cfi_rel_offset x29, 160 75 .cfi_rel_offset x30, 168 76 77 // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs]. 78 str xIP0, [sp] 79 // Place sp in Thread::Current()->top_quick_frame. 80 mov xIP0, sp 81 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 82.endm 83 84 /* 85 * Macro that sets up the callee save frame to conform with 86 * Runtime::CreateCalleeSaveMethod(kRefsOnly). 87 */ 88.macro SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 89 adrp xIP0, :got:_ZN3art7Runtime9instance_E 90 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 91 92 // Our registers aren't intermixed - just spill in order. 93 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 94 95 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefOnly] . 96 // Loads appropriate callee-save-method. 97 ldr xIP0, [xIP0, RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET ] 98 99 sub sp, sp, #96 100 .cfi_adjust_cfa_offset 96 101 102 // Ugly compile-time check, but we only have the preprocessor. 103#if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 96) 104#error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM64) size not as expected." 105#endif 106 107 // GP callee-saves. 108 // x20 paired with ArtMethod* - see below. 109 stp x21, x22, [sp, #16] 110 .cfi_rel_offset x21, 16 111 .cfi_rel_offset x22, 24 112 113 stp x23, x24, [sp, #32] 114 .cfi_rel_offset x23, 32 115 .cfi_rel_offset x24, 40 116 117 stp x25, x26, [sp, #48] 118 .cfi_rel_offset x25, 48 119 .cfi_rel_offset x26, 56 120 121 stp x27, x28, [sp, #64] 122 .cfi_rel_offset x27, 64 123 .cfi_rel_offset x28, 72 124 125 stp x29, xLR, [sp, #80] 126 .cfi_rel_offset x29, 80 127 .cfi_rel_offset x30, 88 128 129 // Store ArtMethod* Runtime::callee_save_methods_[kRefsOnly]. 130 stp xIP0, x20, [sp] 131 .cfi_rel_offset x20, 8 132 133 // Place sp in Thread::Current()->top_quick_frame. 134 mov xIP0, sp 135 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 136.endm 137 138// TODO: Probably no need to restore registers preserved by aapcs64. 139.macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 140 // Callee-saves. 141 ldr x20, [sp, #8] 142 .cfi_restore x20 143 144 ldp x21, x22, [sp, #16] 145 .cfi_restore x21 146 .cfi_restore x22 147 148 ldp x23, x24, [sp, #32] 149 .cfi_restore x23 150 .cfi_restore x24 151 152 ldp x25, x26, [sp, #48] 153 .cfi_restore x25 154 .cfi_restore x26 155 156 ldp x27, x28, [sp, #64] 157 .cfi_restore x27 158 .cfi_restore x28 159 160 ldp x29, xLR, [sp, #80] 161 .cfi_restore x29 162 .cfi_restore x30 163 164 add sp, sp, #96 165 .cfi_adjust_cfa_offset -96 166.endm 167 168.macro POP_REFS_ONLY_CALLEE_SAVE_FRAME 169 add sp, sp, #96 170 .cfi_adjust_cfa_offset - 96 171.endm 172 173.macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 174 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 175 ret 176.endm 177 178 179.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 180 sub sp, sp, #224 181 .cfi_adjust_cfa_offset 224 182 183 // Ugly compile-time check, but we only have the preprocessor. 184#if (FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE != 224) 185#error "REFS_AND_ARGS_CALLEE_SAVE_FRAME(ARM64) size not as expected." 186#endif 187 188 // Stack alignment filler [sp, #8]. 189 // FP args. 190 stp d0, d1, [sp, #16] 191 stp d2, d3, [sp, #32] 192 stp d4, d5, [sp, #48] 193 stp d6, d7, [sp, #64] 194 195 // Core args. 196 stp x1, x2, [sp, #80] 197 .cfi_rel_offset x1, 80 198 .cfi_rel_offset x2, 88 199 200 stp x3, x4, [sp, #96] 201 .cfi_rel_offset x3, 96 202 .cfi_rel_offset x4, 104 203 204 stp x5, x6, [sp, #112] 205 .cfi_rel_offset x5, 112 206 .cfi_rel_offset x6, 120 207 208 // x7, Callee-saves. 209 stp x7, x20, [sp, #128] 210 .cfi_rel_offset x7, 128 211 .cfi_rel_offset x20, 136 212 213 stp x21, x22, [sp, #144] 214 .cfi_rel_offset x21, 144 215 .cfi_rel_offset x22, 152 216 217 stp x23, x24, [sp, #160] 218 .cfi_rel_offset x23, 160 219 .cfi_rel_offset x24, 168 220 221 stp x25, x26, [sp, #176] 222 .cfi_rel_offset x25, 176 223 .cfi_rel_offset x26, 184 224 225 stp x27, x28, [sp, #192] 226 .cfi_rel_offset x27, 192 227 .cfi_rel_offset x28, 200 228 229 // x29(callee-save) and LR. 230 stp x29, xLR, [sp, #208] 231 .cfi_rel_offset x29, 208 232 .cfi_rel_offset x30, 216 233 234.endm 235 236 /* 237 * Macro that sets up the callee save frame to conform with 238 * Runtime::CreateCalleeSaveMethod(kRefsAndArgs). 239 * 240 * TODO This is probably too conservative - saving FP & LR. 241 */ 242.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 243 adrp xIP0, :got:_ZN3art7Runtime9instance_E 244 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 245 246 // Our registers aren't intermixed - just spill in order. 247 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 248 249 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 250 ldr xIP0, [xIP0, RUNTIME_REFS_AND_ARGS_CALLEE_SAVE_FRAME_OFFSET ] 251 252 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 253 254 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 255 // Place sp in Thread::Current()->top_quick_frame. 256 mov xIP0, sp 257 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 258.endm 259 260.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 261 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 262 str x0, [sp, #0] // Store ArtMethod* to bottom of stack. 263 // Place sp in Thread::Current()->top_quick_frame. 264 mov xIP0, sp 265 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 266.endm 267 268// TODO: Probably no need to restore registers preserved by aapcs64. 269.macro RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 270 // FP args. 271 ldp d0, d1, [sp, #16] 272 ldp d2, d3, [sp, #32] 273 ldp d4, d5, [sp, #48] 274 ldp d6, d7, [sp, #64] 275 276 // Core args. 277 ldp x1, x2, [sp, #80] 278 .cfi_restore x1 279 .cfi_restore x2 280 281 ldp x3, x4, [sp, #96] 282 .cfi_restore x3 283 .cfi_restore x4 284 285 ldp x5, x6, [sp, #112] 286 .cfi_restore x5 287 .cfi_restore x6 288 289 // x7, Callee-saves. 290 ldp x7, x20, [sp, #128] 291 .cfi_restore x7 292 .cfi_restore x20 293 294 ldp x21, x22, [sp, #144] 295 .cfi_restore x21 296 .cfi_restore x22 297 298 ldp x23, x24, [sp, #160] 299 .cfi_restore x23 300 .cfi_restore x24 301 302 ldp x25, x26, [sp, #176] 303 .cfi_restore x25 304 .cfi_restore x26 305 306 ldp x27, x28, [sp, #192] 307 .cfi_restore x27 308 .cfi_restore x28 309 310 // x29(callee-save) and LR. 311 ldp x29, xLR, [sp, #208] 312 .cfi_restore x29 313 .cfi_restore x30 314 315 add sp, sp, #224 316 .cfi_adjust_cfa_offset -224 317.endm 318 319.macro RETURN_IF_RESULT_IS_ZERO 320 cbnz x0, 1f // result non-zero branch over 321 ret // return 3221: 323.endm 324 325.macro RETURN_IF_RESULT_IS_NON_ZERO 326 cbz x0, 1f // result zero branch over 327 ret // return 3281: 329.endm 330 331 /* 332 * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending 333 * exception is Thread::Current()->exception_ 334 */ 335.macro DELIVER_PENDING_EXCEPTION 336 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 337 mov x0, xSELF 338 339 // Point of no return. 340 b artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*) 341 brk 0 // Unreached 342.endm 343 344.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 345 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 346 cbnz \reg, 1f 347 ret 3481: 349 DELIVER_PENDING_EXCEPTION 350.endm 351 352.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 353 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 354.endm 355 356// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 357.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 358 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 359.endm 360 361.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 362 cbnz w0, 1f // result non-zero branch over 363 ret // return 3641: 365 DELIVER_PENDING_EXCEPTION 366.endm 367 368.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 369 .extern \cxx_name 370ENTRY \c_name 371 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 372 mov x0, xSELF // pass Thread::Current 373 b \cxx_name // \cxx_name(Thread*) 374END \c_name 375.endm 376 377.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 378 .extern \cxx_name 379ENTRY \c_name 380 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context. 381 mov x1, xSELF // pass Thread::Current. 382 b \cxx_name // \cxx_name(arg, Thread*). 383 brk 0 384END \c_name 385.endm 386 387.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 388 .extern \cxx_name 389ENTRY \c_name 390 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 391 mov x2, xSELF // pass Thread::Current 392 b \cxx_name // \cxx_name(arg1, arg2, Thread*) 393 brk 0 394END \c_name 395.endm 396 397 /* 398 * Called by managed code, saves callee saves and then calls artThrowException 399 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 400 */ 401ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 402 403 /* 404 * Called by managed code to create and deliver a NullPointerException. 405 */ 406NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 407 408 /* 409 * Call installed by a signal handler to create and deliver a NullPointerException. 410 */ 411ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal 412 413 /* 414 * Called by managed code to create and deliver an ArithmeticException. 415 */ 416NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode 417 418 /* 419 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 420 * index, arg2 holds limit. 421 */ 422TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 423 424 /* 425 * Called by managed code to create and deliver a StringIndexOutOfBoundsException 426 * as if thrown from a call to String.charAt(). Arg1 holds index, arg2 holds limit. 427 */ 428TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_string_bounds, artThrowStringBoundsFromCode 429 430 /* 431 * Called by managed code to create and deliver a StackOverflowError. 432 */ 433NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 434 435 /* 436 * Called by managed code to create and deliver a NoSuchMethodError. 437 */ 438ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode 439 440 /* 441 * All generated callsites for interface invokes and invocation slow paths will load arguments 442 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 443 * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. 444 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 445 * 446 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 447 * of the target Method* in x0 and method->code_ in x1. 448 * 449 * If unsuccessful, the helper will return null/????. There will be a pending exception in the 450 * thread and we branch to another stub to deliver it. 451 * 452 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 453 * pointing back to the original caller. 454 * 455 * Adapted from ARM32 code. 456 * 457 * Clobbers xIP0. 458 */ 459.macro INVOKE_TRAMPOLINE_BODY cxx_name 460 .extern \cxx_name 461 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // save callee saves in case allocation triggers GC 462 // Helper signature is always 463 // (method_idx, *this_object, *caller_method, *self, sp) 464 465 mov x2, xSELF // pass Thread::Current 466 mov x3, sp 467 bl \cxx_name // (method_idx, this, Thread*, SP) 468 mov xIP0, x1 // save Method*->code_ 469 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 470 cbz x0, 1f // did we find the target? if not go to exception delivery 471 br xIP0 // tail call to target 4721: 473 DELIVER_PENDING_EXCEPTION 474.endm 475.macro INVOKE_TRAMPOLINE c_name, cxx_name 476ENTRY \c_name 477 INVOKE_TRAMPOLINE_BODY \cxx_name 478END \c_name 479.endm 480 481INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 482 483INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 484INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 485INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 486INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 487 488 489.macro INVOKE_STUB_CREATE_FRAME 490 491SAVE_SIZE=15*8 // x4, x5, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 492SAVE_SIZE_AND_METHOD=SAVE_SIZE+8 493 494 495 mov x9, sp // Save stack pointer. 496 .cfi_register sp,x9 497 498 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 499 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 500 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 501 mov sp, x10 // Set new SP. 502 503 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 504 .cfi_def_cfa_register x10 // before this. 505 .cfi_adjust_cfa_offset SAVE_SIZE 506 507 str x28, [x10, #112] 508 .cfi_rel_offset x28, 112 509 510 stp x26, x27, [x10, #96] 511 .cfi_rel_offset x26, 96 512 .cfi_rel_offset x27, 104 513 514 stp x24, x25, [x10, #80] 515 .cfi_rel_offset x24, 80 516 .cfi_rel_offset x25, 88 517 518 stp x22, x23, [x10, #64] 519 .cfi_rel_offset x22, 64 520 .cfi_rel_offset x23, 72 521 522 stp x20, x21, [x10, #48] 523 .cfi_rel_offset x20, 48 524 .cfi_rel_offset x21, 56 525 526 stp x9, x19, [x10, #32] // Save old stack pointer and x19. 527 .cfi_rel_offset sp, 32 528 .cfi_rel_offset x19, 40 529 530 stp x4, x5, [x10, #16] // Save result and shorty addresses. 531 .cfi_rel_offset x4, 16 532 .cfi_rel_offset x5, 24 533 534 stp xFP, xLR, [x10] // Store LR & FP. 535 .cfi_rel_offset x29, 0 536 .cfi_rel_offset x30, 8 537 538 mov xFP, x10 // Use xFP now, as it's callee-saved. 539 .cfi_def_cfa_register x29 540 mov xSELF, x3 // Move thread pointer into SELF register. 541 542 // Copy arguments into stack frame. 543 // Use simple copy routine for now. 544 // 4 bytes per slot. 545 // X1 - source address 546 // W2 - args length 547 // X9 - destination address. 548 // W10 - temporary 549 add x9, sp, #8 // Destination address is bottom of stack + null. 550 551 // Copy parameters into the stack. Use numeric label as this is a macro and Clang's assembler 552 // does not have unique-id variables. 5531: 554 cmp w2, #0 555 beq 2f 556 sub w2, w2, #4 // Need 65536 bytes of range. 557 ldr w10, [x1, x2] 558 str w10, [x9, x2] 559 560 b 1b 561 5622: 563 // Store null into ArtMethod* at bottom of frame. 564 str xzr, [sp] 565.endm 566 567.macro INVOKE_STUB_CALL_AND_RETURN 568 569 // load method-> METHOD_QUICK_CODE_OFFSET 570 ldr x9, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 571 // Branch to method. 572 blr x9 573 574 // Restore return value address and shorty address. 575 ldp x4,x5, [xFP, #16] 576 .cfi_restore x4 577 .cfi_restore x5 578 579 ldr x28, [xFP, #112] 580 .cfi_restore x28 581 582 ldp x26, x27, [xFP, #96] 583 .cfi_restore x26 584 .cfi_restore x27 585 586 ldp x24, x25, [xFP, #80] 587 .cfi_restore x24 588 .cfi_restore x25 589 590 ldp x22, x23, [xFP, #64] 591 .cfi_restore x22 592 .cfi_restore x23 593 594 ldp x20, x21, [xFP, #48] 595 .cfi_restore x20 596 .cfi_restore x21 597 598 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 599 ldrb w10, [x5] 600 601 // Check the return type and store the correct register into the jvalue in memory. 602 // Use numeric label as this is a macro and Clang's assembler does not have unique-id variables. 603 604 // Don't set anything for a void type. 605 cmp w10, #'V' 606 beq 3f 607 608 // Is it a double? 609 cmp w10, #'D' 610 bne 1f 611 str d0, [x4] 612 b 3f 613 6141: // Is it a float? 615 cmp w10, #'F' 616 bne 2f 617 str s0, [x4] 618 b 3f 619 6202: // Just store x0. Doesn't matter if it is 64 or 32 bits. 621 str x0, [x4] 622 6233: // Finish up. 624 ldp x2, x19, [xFP, #32] // Restore stack pointer and x19. 625 .cfi_restore x19 626 mov sp, x2 627 .cfi_restore sp 628 629 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 630 .cfi_restore x29 631 .cfi_restore x30 632 633 ret 634 635.endm 636 637 638/* 639 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 640 * uint32_t *args, x1 641 * uint32_t argsize, w2 642 * Thread *self, x3 643 * JValue *result, x4 644 * char *shorty); x5 645 * +----------------------+ 646 * | | 647 * | C/C++ frame | 648 * | LR'' | 649 * | FP'' | <- SP' 650 * +----------------------+ 651 * +----------------------+ 652 * | x28 | <- TODO: Remove callee-saves. 653 * | : | 654 * | x19 | 655 * | SP' | 656 * | X5 | 657 * | X4 | Saved registers 658 * | LR' | 659 * | FP' | <- FP 660 * +----------------------+ 661 * | uint32_t out[n-1] | 662 * | : : | Outs 663 * | uint32_t out[0] | 664 * | ArtMethod* | <- SP value=null 665 * +----------------------+ 666 * 667 * Outgoing registers: 668 * x0 - Method* 669 * x1-x7 - integer parameters. 670 * d0-d7 - Floating point parameters. 671 * xSELF = self 672 * SP = & of ArtMethod* 673 * x1 = "this" pointer. 674 * 675 */ 676ENTRY art_quick_invoke_stub 677 // Spill registers as per AACPS64 calling convention. 678 INVOKE_STUB_CREATE_FRAME 679 680 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 681 // Parse the passed shorty to determine which register to load. 682 // Load addresses for routines that load WXSD registers. 683 adr x11, .LstoreW2 684 adr x12, .LstoreX2 685 adr x13, .LstoreS0 686 adr x14, .LstoreD0 687 688 // Initialize routine offsets to 0 for integers and floats. 689 // x8 for integers, x15 for floating point. 690 mov x8, #0 691 mov x15, #0 692 693 add x10, x5, #1 // Load shorty address, plus one to skip return value. 694 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 695 696 // Loop to fill registers. 697.LfillRegisters: 698 ldrb w17, [x10], #1 // Load next character in signature, and increment. 699 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 700 701 cmp w17, #'F' // is this a float? 702 bne .LisDouble 703 704 cmp x15, # 8*12 // Skip this load if all registers full. 705 beq .Ladvance4 706 707 add x17, x13, x15 // Calculate subroutine to jump to. 708 br x17 709 710.LisDouble: 711 cmp w17, #'D' // is this a double? 712 bne .LisLong 713 714 cmp x15, # 8*12 // Skip this load if all registers full. 715 beq .Ladvance8 716 717 add x17, x14, x15 // Calculate subroutine to jump to. 718 br x17 719 720.LisLong: 721 cmp w17, #'J' // is this a long? 722 bne .LisOther 723 724 cmp x8, # 6*12 // Skip this load if all registers full. 725 beq .Ladvance8 726 727 add x17, x12, x8 // Calculate subroutine to jump to. 728 br x17 729 730.LisOther: // Everything else takes one vReg. 731 cmp x8, # 6*12 // Skip this load if all registers full. 732 beq .Ladvance4 733 734 add x17, x11, x8 // Calculate subroutine to jump to. 735 br x17 736 737.Ladvance4: 738 add x9, x9, #4 739 b .LfillRegisters 740 741.Ladvance8: 742 add x9, x9, #8 743 b .LfillRegisters 744 745// Macro for loading a parameter into a register. 746// counter - the register with offset into these tables 747// size - the size of the register - 4 or 8 bytes. 748// register - the name of the register to be loaded. 749.macro LOADREG counter size register return 750 ldr \register , [x9], #\size 751 add \counter, \counter, 12 752 b \return 753.endm 754 755// Store ints. 756.LstoreW2: 757 LOADREG x8 4 w2 .LfillRegisters 758 LOADREG x8 4 w3 .LfillRegisters 759 LOADREG x8 4 w4 .LfillRegisters 760 LOADREG x8 4 w5 .LfillRegisters 761 LOADREG x8 4 w6 .LfillRegisters 762 LOADREG x8 4 w7 .LfillRegisters 763 764// Store longs. 765.LstoreX2: 766 LOADREG x8 8 x2 .LfillRegisters 767 LOADREG x8 8 x3 .LfillRegisters 768 LOADREG x8 8 x4 .LfillRegisters 769 LOADREG x8 8 x5 .LfillRegisters 770 LOADREG x8 8 x6 .LfillRegisters 771 LOADREG x8 8 x7 .LfillRegisters 772 773// Store singles. 774.LstoreS0: 775 LOADREG x15 4 s0 .LfillRegisters 776 LOADREG x15 4 s1 .LfillRegisters 777 LOADREG x15 4 s2 .LfillRegisters 778 LOADREG x15 4 s3 .LfillRegisters 779 LOADREG x15 4 s4 .LfillRegisters 780 LOADREG x15 4 s5 .LfillRegisters 781 LOADREG x15 4 s6 .LfillRegisters 782 LOADREG x15 4 s7 .LfillRegisters 783 784// Store doubles. 785.LstoreD0: 786 LOADREG x15 8 d0 .LfillRegisters 787 LOADREG x15 8 d1 .LfillRegisters 788 LOADREG x15 8 d2 .LfillRegisters 789 LOADREG x15 8 d3 .LfillRegisters 790 LOADREG x15 8 d4 .LfillRegisters 791 LOADREG x15 8 d5 .LfillRegisters 792 LOADREG x15 8 d6 .LfillRegisters 793 LOADREG x15 8 d7 .LfillRegisters 794 795 796.LcallFunction: 797 798 INVOKE_STUB_CALL_AND_RETURN 799 800END art_quick_invoke_stub 801 802/* extern"C" 803 * void art_quick_invoke_static_stub(ArtMethod *method, x0 804 * uint32_t *args, x1 805 * uint32_t argsize, w2 806 * Thread *self, x3 807 * JValue *result, x4 808 * char *shorty); x5 809 */ 810ENTRY art_quick_invoke_static_stub 811 // Spill registers as per AACPS64 calling convention. 812 INVOKE_STUB_CREATE_FRAME 813 814 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 815 // Parse the passed shorty to determine which register to load. 816 // Load addresses for routines that load WXSD registers. 817 adr x11, .LstoreW1_2 818 adr x12, .LstoreX1_2 819 adr x13, .LstoreS0_2 820 adr x14, .LstoreD0_2 821 822 // Initialize routine offsets to 0 for integers and floats. 823 // x8 for integers, x15 for floating point. 824 mov x8, #0 825 mov x15, #0 826 827 add x10, x5, #1 // Load shorty address, plus one to skip return value. 828 829 // Loop to fill registers. 830.LfillRegisters2: 831 ldrb w17, [x10], #1 // Load next character in signature, and increment. 832 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 833 834 cmp w17, #'F' // is this a float? 835 bne .LisDouble2 836 837 cmp x15, # 8*12 // Skip this load if all registers full. 838 beq .Ladvance4_2 839 840 add x17, x13, x15 // Calculate subroutine to jump to. 841 br x17 842 843.LisDouble2: 844 cmp w17, #'D' // is this a double? 845 bne .LisLong2 846 847 cmp x15, # 8*12 // Skip this load if all registers full. 848 beq .Ladvance8_2 849 850 add x17, x14, x15 // Calculate subroutine to jump to. 851 br x17 852 853.LisLong2: 854 cmp w17, #'J' // is this a long? 855 bne .LisOther2 856 857 cmp x8, # 7*12 // Skip this load if all registers full. 858 beq .Ladvance8_2 859 860 add x17, x12, x8 // Calculate subroutine to jump to. 861 br x17 862 863.LisOther2: // Everything else takes one vReg. 864 cmp x8, # 7*12 // Skip this load if all registers full. 865 beq .Ladvance4_2 866 867 add x17, x11, x8 // Calculate subroutine to jump to. 868 br x17 869 870.Ladvance4_2: 871 add x9, x9, #4 872 b .LfillRegisters2 873 874.Ladvance8_2: 875 add x9, x9, #8 876 b .LfillRegisters2 877 878// Store ints. 879.LstoreW1_2: 880 LOADREG x8 4 w1 .LfillRegisters2 881 LOADREG x8 4 w2 .LfillRegisters2 882 LOADREG x8 4 w3 .LfillRegisters2 883 LOADREG x8 4 w4 .LfillRegisters2 884 LOADREG x8 4 w5 .LfillRegisters2 885 LOADREG x8 4 w6 .LfillRegisters2 886 LOADREG x8 4 w7 .LfillRegisters2 887 888// Store longs. 889.LstoreX1_2: 890 LOADREG x8 8 x1 .LfillRegisters2 891 LOADREG x8 8 x2 .LfillRegisters2 892 LOADREG x8 8 x3 .LfillRegisters2 893 LOADREG x8 8 x4 .LfillRegisters2 894 LOADREG x8 8 x5 .LfillRegisters2 895 LOADREG x8 8 x6 .LfillRegisters2 896 LOADREG x8 8 x7 .LfillRegisters2 897 898// Store singles. 899.LstoreS0_2: 900 LOADREG x15 4 s0 .LfillRegisters2 901 LOADREG x15 4 s1 .LfillRegisters2 902 LOADREG x15 4 s2 .LfillRegisters2 903 LOADREG x15 4 s3 .LfillRegisters2 904 LOADREG x15 4 s4 .LfillRegisters2 905 LOADREG x15 4 s5 .LfillRegisters2 906 LOADREG x15 4 s6 .LfillRegisters2 907 LOADREG x15 4 s7 .LfillRegisters2 908 909// Store doubles. 910.LstoreD0_2: 911 LOADREG x15 8 d0 .LfillRegisters2 912 LOADREG x15 8 d1 .LfillRegisters2 913 LOADREG x15 8 d2 .LfillRegisters2 914 LOADREG x15 8 d3 .LfillRegisters2 915 LOADREG x15 8 d4 .LfillRegisters2 916 LOADREG x15 8 d5 .LfillRegisters2 917 LOADREG x15 8 d6 .LfillRegisters2 918 LOADREG x15 8 d7 .LfillRegisters2 919 920 921.LcallFunction2: 922 923 INVOKE_STUB_CALL_AND_RETURN 924 925END art_quick_invoke_static_stub 926 927 928 929/* extern"C" void art_quick_osr_stub(void** stack, x0 930 * size_t stack_size_in_bytes, x1 931 * const uin8_t* native_pc, x2 932 * JValue *result, x3 933 * char *shorty, x4 934 * Thread *self) x5 935 */ 936ENTRY art_quick_osr_stub 937SAVE_SIZE=15*8 // x3, x4, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 938 mov x9, sp // Save stack pointer. 939 .cfi_register sp,x9 940 941 sub x10, sp, # SAVE_SIZE 942 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 943 mov sp, x10 // Set new SP. 944 945 str x28, [sp, #112] 946 stp x26, x27, [sp, #96] 947 stp x24, x25, [sp, #80] 948 stp x22, x23, [sp, #64] 949 stp x20, x21, [sp, #48] 950 stp x9, x19, [sp, #32] // Save old stack pointer and x19. 951 stp x3, x4, [sp, #16] // Save result and shorty addresses. 952 stp xFP, xLR, [sp] // Store LR & FP. 953 mov xSELF, x5 // Move thread pointer into SELF register. 954 955 sub sp, sp, #16 956 str xzr, [sp] // Store null for ArtMethod* slot 957 // Branch to stub. 958 bl .Losr_entry 959 add sp, sp, #16 960 961 // Restore return value address and shorty address. 962 ldp x3,x4, [sp, #16] 963 ldr x28, [sp, #112] 964 ldp x26, x27, [sp, #96] 965 ldp x24, x25, [sp, #80] 966 ldp x22, x23, [sp, #64] 967 ldp x20, x21, [sp, #48] 968 969 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 970 ldrb w10, [x4] 971 972 // Check the return type and store the correct register into the jvalue in memory. 973 974 // Don't set anything for a void type. 975 cmp w10, #'V' 976 beq .Losr_exit 977 978 // Is it a double? 979 cmp w10, #'D' 980 bne .Lno_double 981 str d0, [x3] 982 b .Losr_exit 983 984.Lno_double: // Is it a float? 985 cmp w10, #'F' 986 bne .Lno_float 987 str s0, [x3] 988 b .Losr_exit 989 990.Lno_float: // Just store x0. Doesn't matter if it is 64 or 32 bits. 991 str x0, [x3] 992 993.Losr_exit: // Finish up. 994 ldp x2, x19, [sp, #32] // Restore stack pointer and x19. 995 ldp xFP, xLR, [sp] // Restore old frame pointer and link register. 996 mov sp, x2 997 ret 998 999.Losr_entry: 1000 // Update stack pointer for the callee 1001 sub sp, sp, x1 1002 1003 // Update link register slot expected by the callee. 1004 sub w1, w1, #8 1005 str lr, [sp, x1] 1006 1007 // Copy arguments into stack frame. 1008 // Use simple copy routine for now. 1009 // 4 bytes per slot. 1010 // X0 - source address 1011 // W1 - args length 1012 // SP - destination address. 1013 // W10 - temporary 1014.Losr_loop_entry: 1015 cmp w1, #0 1016 beq .Losr_loop_exit 1017 sub w1, w1, #4 1018 ldr w10, [x0, x1] 1019 str w10, [sp, x1] 1020 b .Losr_loop_entry 1021 1022.Losr_loop_exit: 1023 // Branch to the OSR entry point. 1024 br x2 1025 1026END art_quick_osr_stub 1027 1028 /* 1029 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 1030 */ 1031 1032ENTRY art_quick_do_long_jump 1033 // Load FPRs 1034 ldp d0, d1, [x1], #16 1035 ldp d2, d3, [x1], #16 1036 ldp d4, d5, [x1], #16 1037 ldp d6, d7, [x1], #16 1038 ldp d8, d9, [x1], #16 1039 ldp d10, d11, [x1], #16 1040 ldp d12, d13, [x1], #16 1041 ldp d14, d15, [x1], #16 1042 ldp d16, d17, [x1], #16 1043 ldp d18, d19, [x1], #16 1044 ldp d20, d21, [x1], #16 1045 ldp d22, d23, [x1], #16 1046 ldp d24, d25, [x1], #16 1047 ldp d26, d27, [x1], #16 1048 ldp d28, d29, [x1], #16 1049 ldp d30, d31, [x1] 1050 1051 // Load GPRs 1052 // TODO: lots of those are smashed, could optimize. 1053 add x0, x0, #30*8 1054 ldp x30, x1, [x0], #-16 // LR & SP 1055 ldp x28, x29, [x0], #-16 1056 ldp x26, x27, [x0], #-16 1057 ldp x24, x25, [x0], #-16 1058 ldp x22, x23, [x0], #-16 1059 ldp x20, x21, [x0], #-16 1060 ldp x18, x19, [x0], #-16 1061 ldp x16, x17, [x0], #-16 1062 ldp x14, x15, [x0], #-16 1063 ldp x12, x13, [x0], #-16 1064 ldp x10, x11, [x0], #-16 1065 ldp x8, x9, [x0], #-16 1066 ldp x6, x7, [x0], #-16 1067 ldp x4, x5, [x0], #-16 1068 ldp x2, x3, [x0], #-16 1069 mov sp, x1 1070 1071 // Need to load PC, it's at the end (after the space for the unused XZR). Use x1. 1072 ldr x1, [x0, #33*8] 1073 // And the value of x0. 1074 ldr x0, [x0] 1075 1076 br x1 1077END art_quick_do_long_jump 1078 1079 /* 1080 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 1081 * possibly null object to lock. 1082 * 1083 * Derived from arm32 code. 1084 */ 1085 .extern artLockObjectFromCode 1086ENTRY art_quick_lock_object 1087 cbz w0, .Lslow_lock 1088 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1089.Lretry_lock: 1090 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 1091 ldxr w1, [x4] 1092 mov x3, x1 1093 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1094 cbnz w3, .Lnot_unlocked // already thin locked 1095 // unlocked case - x1: original lock word that's zero except for the read barrier bits. 1096 orr x2, x1, x2 // x2 holds thread id with count of 0 with preserved read barrier bits 1097 stxr w3, w2, [x4] 1098 cbnz w3, .Llock_stxr_fail // store failed, retry 1099 dmb ishld // full (LoadLoad|LoadStore) memory barrier 1100 ret 1101.Lnot_unlocked: // x1: original lock word 1102 lsr w3, w1, LOCK_WORD_STATE_SHIFT 1103 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 1104 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1105 uxth w2, w2 // zero top 16 bits 1106 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 1107 // else contention, go to slow path 1108 mov x3, x1 // copy the lock word to check count overflow. 1109 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits. 1110 add w2, w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count in lock word placing in w2 to check overflow 1111 lsr w3, w2, LOCK_WORD_READ_BARRIER_STATE_SHIFT // if either of the upper two bits (28-29) are set, we overflowed. 1112 cbnz w3, .Lslow_lock // if we overflow the count go slow path 1113 add w2, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count for real 1114 stxr w3, w2, [x4] 1115 cbnz w3, .Llock_stxr_fail // store failed, retry 1116 ret 1117.Llock_stxr_fail: 1118 b .Lretry_lock // retry 1119.Lslow_lock: 1120 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case we block 1121 mov x1, xSELF // pass Thread::Current 1122 bl artLockObjectFromCode // (Object* obj, Thread*) 1123 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1124 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1125END art_quick_lock_object 1126 1127ENTRY art_quick_lock_object_no_inline 1128 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case we block 1129 mov x1, xSELF // pass Thread::Current 1130 bl artLockObjectFromCode // (Object* obj, Thread*) 1131 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1132 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1133END art_quick_lock_object_no_inline 1134 1135 /* 1136 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1137 * x0 holds the possibly null object to lock. 1138 * 1139 * Derived from arm32 code. 1140 */ 1141 .extern artUnlockObjectFromCode 1142ENTRY art_quick_unlock_object 1143 cbz x0, .Lslow_unlock 1144 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1145.Lretry_unlock: 1146#ifndef USE_READ_BARRIER 1147 ldr w1, [x4] 1148#else 1149 ldxr w1, [x4] // Need to use atomic instructions for read barrier 1150#endif 1151 lsr w2, w1, LOCK_WORD_STATE_SHIFT 1152 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1153 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1154 mov x3, x1 // copy lock word to check thread id equality 1155 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1156 eor w3, w3, w2 // lock_word.ThreadId() ^ self->ThreadId() 1157 uxth w3, w3 // zero top 16 bits 1158 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1159 mov x3, x1 // copy lock word to detect transition to unlocked 1160 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1161 cmp w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE 1162 bpl .Lrecursive_thin_unlock 1163 // transition to unlocked 1164 mov x3, x1 1165 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK // w3: zero except for the preserved read barrier bits 1166 dmb ish // full (LoadStore|StoreStore) memory barrier 1167#ifndef USE_READ_BARRIER 1168 str w3, [x4] 1169#else 1170 stxr w2, w3, [x4] // Need to use atomic instructions for read barrier 1171 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1172#endif 1173 ret 1174.Lrecursive_thin_unlock: // w1: original lock word 1175 sub w1, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // decrement count 1176#ifndef USE_READ_BARRIER 1177 str w1, [x4] 1178#else 1179 stxr w2, w1, [x4] // Need to use atomic instructions for read barrier 1180 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1181#endif 1182 ret 1183.Lunlock_stxr_fail: 1184 b .Lretry_unlock // retry 1185.Lslow_unlock: 1186 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case exception allocation triggers GC 1187 mov x1, xSELF // pass Thread::Current 1188 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1189 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1190 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1191END art_quick_unlock_object 1192 1193ENTRY art_quick_unlock_object_no_inline 1194 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case exception allocation triggers GC 1195 mov x1, xSELF // pass Thread::Current 1196 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1197 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1198 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1199END art_quick_unlock_object_no_inline 1200 1201 /* 1202 * Entry from managed code that calls artIsAssignableFromCode and on failure calls 1203 * artThrowClassCastException. 1204 */ 1205 .extern artThrowClassCastException 1206ENTRY art_quick_check_cast 1207 // Store arguments and link register 1208 // Stack needs to be 16B aligned on calls. 1209 stp x0, x1, [sp,#-32]! 1210 .cfi_adjust_cfa_offset 32 1211 .cfi_rel_offset x0, 0 1212 .cfi_rel_offset x1, 8 1213 str xLR, [sp, #24] 1214 .cfi_rel_offset x30, 24 1215 1216 // Call runtime code 1217 bl artIsAssignableFromCode 1218 1219 // Check for exception 1220 cbz x0, .Lthrow_class_cast_exception 1221 1222 // Restore and return 1223 ldr xLR, [sp, #24] 1224 .cfi_restore x30 1225 ldp x0, x1, [sp], #32 1226 .cfi_restore x0 1227 .cfi_restore x1 1228 .cfi_adjust_cfa_offset -32 1229 ret 1230 1231 .cfi_adjust_cfa_offset 32 // Reset unwind info so following code unwinds. 1232 1233.Lthrow_class_cast_exception: 1234 // Restore 1235 ldr xLR, [sp, #24] 1236 .cfi_restore x30 1237 ldp x0, x1, [sp], #32 1238 .cfi_restore x0 1239 .cfi_restore x1 1240 .cfi_adjust_cfa_offset -32 1241 1242 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 1243 mov x2, xSELF // pass Thread::Current 1244 b artThrowClassCastException // (Class*, Class*, Thread*) 1245 brk 0 // We should not return here... 1246END art_quick_check_cast 1247 1248// Restore xReg's value from [sp, #offset] if xReg is not the same as xExclude. 1249.macro POP_REG_NE xReg, offset, xExclude 1250 .ifnc \xReg, \xExclude 1251 ldr \xReg, [sp, #\offset] // restore xReg 1252 .cfi_restore \xReg 1253 .endif 1254.endm 1255 1256// Restore xReg1's value from [sp, #offset] if xReg1 is not the same as xExclude. 1257// Restore xReg2's value from [sp, #(offset + 8)] if xReg2 is not the same as xExclude. 1258.macro POP_REGS_NE xReg1, xReg2, offset, xExclude 1259 .ifc \xReg1, \xExclude 1260 ldr \xReg2, [sp, #(\offset + 8)] // restore xReg2 1261 .else 1262 .ifc \xReg2, \xExclude 1263 ldr \xReg1, [sp, #\offset] // restore xReg1 1264 .else 1265 ldp \xReg1, \xReg2, [sp, #\offset] // restore xReg1 and xReg2 1266 .endif 1267 .endif 1268 .cfi_restore \xReg1 1269 .cfi_restore \xReg2 1270.endm 1271 1272 /* 1273 * Macro to insert read barrier, only used in art_quick_aput_obj. 1274 * xDest, wDest and xObj are registers, offset is a defined literal such as 1275 * MIRROR_OBJECT_CLASS_OFFSET. Dest needs both x and w versions of the same register to handle 1276 * name mismatch between instructions. This macro uses the lower 32b of register when possible. 1277 * TODO: When read barrier has a fast path, add heap unpoisoning support for the fast path. 1278 */ 1279.macro READ_BARRIER xDest, wDest, xObj, offset 1280#ifdef USE_READ_BARRIER 1281 // Store registers used in art_quick_aput_obj (x0-x4, LR), stack is 16B aligned. 1282 stp x0, x1, [sp, #-48]! 1283 .cfi_adjust_cfa_offset 48 1284 .cfi_rel_offset x0, 0 1285 .cfi_rel_offset x1, 8 1286 stp x2, x3, [sp, #16] 1287 .cfi_rel_offset x2, 16 1288 .cfi_rel_offset x3, 24 1289 stp x4, xLR, [sp, #32] 1290 .cfi_rel_offset x4, 32 1291 .cfi_rel_offset x30, 40 1292 1293 // mov x0, \xRef // pass ref in x0 (no-op for now since parameter ref is unused) 1294 .ifnc \xObj, x1 1295 mov x1, \xObj // pass xObj 1296 .endif 1297 mov w2, #\offset // pass offset 1298 bl artReadBarrierSlow // artReadBarrierSlow(ref, xObj, offset) 1299 // No need to unpoison return value in w0, artReadBarrierSlow() would do the unpoisoning. 1300 .ifnc \wDest, w0 1301 mov \wDest, w0 // save return value in wDest 1302 .endif 1303 1304 // Conditionally restore saved registers 1305 POP_REG_NE x0, 0, \xDest 1306 POP_REG_NE x1, 8, \xDest 1307 POP_REG_NE x2, 16, \xDest 1308 POP_REG_NE x3, 24, \xDest 1309 POP_REG_NE x4, 32, \xDest 1310 ldr xLR, [sp, #40] 1311 .cfi_restore x30 1312 add sp, sp, #48 1313 .cfi_adjust_cfa_offset -48 1314#else 1315 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1316 UNPOISON_HEAP_REF \wDest 1317#endif // USE_READ_BARRIER 1318.endm 1319 1320 /* 1321 * Entry from managed code for array put operations of objects where the value being stored 1322 * needs to be checked for compatibility. 1323 * x0 = array, x1 = index, x2 = value 1324 * 1325 * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We 1326 * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by 1327 * using index-zero-extension in load/stores. 1328 * 1329 * Temporaries: x3, x4 1330 * TODO: x4 OK? ip seems wrong here. 1331 */ 1332ENTRY art_quick_aput_obj_with_null_and_bound_check 1333 tst x0, x0 1334 bne art_quick_aput_obj_with_bound_check 1335 b art_quick_throw_null_pointer_exception 1336END art_quick_aput_obj_with_null_and_bound_check 1337 1338ENTRY art_quick_aput_obj_with_bound_check 1339 ldr w3, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] 1340 cmp w3, w1 1341 bhi art_quick_aput_obj 1342 mov x0, x1 1343 mov x1, x3 1344 b art_quick_throw_array_bounds 1345END art_quick_aput_obj_with_bound_check 1346 1347#ifdef USE_READ_BARRIER 1348 .extern artReadBarrierSlow 1349#endif 1350ENTRY art_quick_aput_obj 1351 cbz x2, .Ldo_aput_null 1352 READ_BARRIER x3, w3, x0, MIRROR_OBJECT_CLASS_OFFSET // Heap reference = 32b 1353 // This also zero-extends to x3 1354 READ_BARRIER x4, w4, x2, MIRROR_OBJECT_CLASS_OFFSET // Heap reference = 32b 1355 // This also zero-extends to x4 1356 READ_BARRIER x3, w3, x3, MIRROR_CLASS_COMPONENT_TYPE_OFFSET // Heap reference = 32b 1357 // This also zero-extends to x3 1358 cmp w3, w4 // value's type == array's component type - trivial assignability 1359 bne .Lcheck_assignability 1360.Ldo_aput: 1361 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1362 // "Compress" = do nothing 1363 POISON_HEAP_REF w2 1364 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1365 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1366 lsr x0, x0, #7 1367 strb w3, [x3, x0] 1368 ret 1369.Ldo_aput_null: 1370 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1371 // "Compress" = do nothing 1372 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1373 ret 1374.Lcheck_assignability: 1375 // Store arguments and link register 1376 stp x0, x1, [sp,#-32]! 1377 .cfi_adjust_cfa_offset 32 1378 .cfi_rel_offset x0, 0 1379 .cfi_rel_offset x1, 8 1380 stp x2, xLR, [sp, #16] 1381 .cfi_rel_offset x2, 16 1382 .cfi_rel_offset x30, 24 1383 1384 // Call runtime code 1385 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1386 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1387 bl artIsAssignableFromCode 1388 1389 // Check for exception 1390 cbz x0, .Lthrow_array_store_exception 1391 1392 // Restore 1393 ldp x2, x30, [sp, #16] 1394 .cfi_restore x2 1395 .cfi_restore x30 1396 ldp x0, x1, [sp], #32 1397 .cfi_restore x0 1398 .cfi_restore x1 1399 .cfi_adjust_cfa_offset -32 1400 1401 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1402 // "Compress" = do nothing 1403 POISON_HEAP_REF w2 1404 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1405 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1406 lsr x0, x0, #7 1407 strb w3, [x3, x0] 1408 ret 1409 .cfi_adjust_cfa_offset 32 // 4 restores after cbz for unwinding. 1410.Lthrow_array_store_exception: 1411 ldp x2, x30, [sp, #16] 1412 .cfi_restore x2 1413 .cfi_restore x30 1414 ldp x0, x1, [sp], #32 1415 .cfi_restore x0 1416 .cfi_restore x1 1417 .cfi_adjust_cfa_offset -32 1418 1419 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1420 mov x1, x2 // Pass value. 1421 mov x2, xSELF // Pass Thread::Current. 1422 b artThrowArrayStoreException // (Object*, Object*, Thread*). 1423 brk 0 // Unreached. 1424END art_quick_aput_obj 1425 1426// Macro to facilitate adding new allocation entrypoints. 1427.macro ONE_ARG_DOWNCALL name, entrypoint, return 1428 .extern \entrypoint 1429ENTRY \name 1430 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1431 mov x1, xSELF // pass Thread::Current 1432 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1433 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1434 \return 1435END \name 1436.endm 1437 1438// Macro to facilitate adding new allocation entrypoints. 1439.macro TWO_ARG_DOWNCALL name, entrypoint, return 1440 .extern \entrypoint 1441ENTRY \name 1442 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1443 mov x2, xSELF // pass Thread::Current 1444 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1445 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1446 \return 1447END \name 1448.endm 1449 1450// Macro to facilitate adding new allocation entrypoints. 1451.macro THREE_ARG_DOWNCALL name, entrypoint, return 1452 .extern \entrypoint 1453ENTRY \name 1454 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1455 mov x3, xSELF // pass Thread::Current 1456 bl \entrypoint 1457 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1458 \return 1459END \name 1460.endm 1461 1462// Macro to facilitate adding new allocation entrypoints. 1463.macro FOUR_ARG_DOWNCALL name, entrypoint, return 1464 .extern \entrypoint 1465ENTRY \name 1466 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1467 mov x4, xSELF // pass Thread::Current 1468 bl \entrypoint // 1469 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1470 \return 1471 DELIVER_PENDING_EXCEPTION 1472END \name 1473.endm 1474 1475// Macros taking opportunity of code similarities for downcalls with referrer. 1476.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1477 .extern \entrypoint 1478ENTRY \name 1479 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1480 ldr x1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1481 mov x2, xSELF // pass Thread::Current 1482 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1483 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1484 \return 1485END \name 1486.endm 1487 1488.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1489 .extern \entrypoint 1490ENTRY \name 1491 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1492 ldr x2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1493 mov x3, xSELF // pass Thread::Current 1494 bl \entrypoint 1495 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1496 \return 1497END \name 1498.endm 1499 1500.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1501 .extern \entrypoint 1502ENTRY \name 1503 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1504 ldr x3, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1505 mov x4, xSELF // pass Thread::Current 1506 bl \entrypoint 1507 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1508 \return 1509END \name 1510.endm 1511 1512.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1513 cbz w0, 1f // result zero branch over 1514 ret // return 15151: 1516 DELIVER_PENDING_EXCEPTION 1517.endm 1518 1519 /* 1520 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 1521 * failure. 1522 */ 1523TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1524 1525 /* 1526 * Entry from managed code when uninitialized static storage, this stub will run the class 1527 * initializer and deliver the exception on error. On success the static storage base is 1528 * returned. 1529 */ 1530ONE_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1531 1532ONE_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1533ONE_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1534 1535ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1536ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1537ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1538ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1539ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1540ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1541ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1542 1543TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1544TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1545TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1546TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1547TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1548TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1549TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1550 1551TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1552TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1553TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1554TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1555 1556THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1557THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1558THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1559THREE_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1560THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1561 1562// This is separated out as the argument order is different. 1563 .extern artSet64StaticFromCode 1564ENTRY art_quick_set64_static 1565 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1566 ldr x1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1567 // x2 contains the parameter 1568 mov x3, xSELF // pass Thread::Current 1569 bl artSet64StaticFromCode 1570 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1571 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1572END art_quick_set64_static 1573 1574 /* 1575 * Entry from managed code to resolve a string, this stub will allocate a String and deliver an 1576 * exception on error. On success the String is returned. w0 holds the string index. The fast 1577 * path check for hit in strings cache has already been performed. 1578 */ 1579ONE_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1580 1581// Generate the allocation entrypoints for each allocator. 1582GENERATE_ALLOC_ENTRYPOINTS_FOR_EACH_ALLOCATOR 1583 1584// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_rosalloc, RosAlloc). 1585ENTRY art_quick_alloc_object_rosalloc 1586 // Fast path rosalloc allocation. 1587 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 1588 // x2-x7: free. 1589 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 1590 // Load the class (x2) 1591 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 1592 cbz x2, .Lart_quick_alloc_object_rosalloc_slow_path // Check null class 1593 // Check class status. 1594 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] 1595 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 1596 bne .Lart_quick_alloc_object_rosalloc_slow_path 1597 // Add a fake dependence from the 1598 // following access flag and size 1599 // loads to the status load. 1600 // This is to prevent those loads 1601 // from being reordered above the 1602 // status load and reading wrong 1603 // values (an alternative is to use 1604 // a load-acquire for the status). 1605 eor x3, x3, x3 1606 add x2, x2, x3 1607 // Check access flags has 1608 // kAccClassIsFinalizable 1609 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 1610 tst x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE 1611 bne .Lart_quick_alloc_object_rosalloc_slow_path 1612 ldr x3, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Check if the thread local 1613 // allocation stack has room. 1614 // ldp won't work due to large offset. 1615 ldr x4, [xSELF, #THREAD_LOCAL_ALLOC_STACK_END_OFFSET] 1616 cmp x3, x4 1617 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1618 ldr w3, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x3) 1619 cmp x3, #ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE // Check if the size is for a thread 1620 // local allocation 1621 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1622 // Compute the rosalloc bracket index 1623 // from the size. 1624 // Align up the size by the rosalloc 1625 // bracket quantum size and divide 1626 // by the quantum size and subtract 1627 // by 1. This code is a shorter but 1628 // equivalent version. 1629 sub x3, x3, #1 1630 lsr x3, x3, #ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT 1631 // Load the rosalloc run (x4) 1632 add x4, xSELF, x3, lsl #POINTER_SIZE_SHIFT 1633 ldr x4, [x4, #THREAD_ROSALLOC_RUNS_OFFSET] 1634 // Load the free list head (x3). This 1635 // will be the return val. 1636 ldr x3, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1637 cbz x3, .Lart_quick_alloc_object_rosalloc_slow_path 1638 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1639 ldr x1, [x3, #ROSALLOC_SLOT_NEXT_OFFSET] // Load the next pointer of the head 1640 // and update the list head with the 1641 // next pointer. 1642 str x1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1643 // Store the class pointer in the 1644 // header. This also overwrites the 1645 // next pointer. The offsets are 1646 // asserted to match. 1647#if ROSALLOC_SLOT_NEXT_OFFSET != MIRROR_OBJECT_CLASS_OFFSET 1648#error "Class pointer needs to overwrite next pointer." 1649#endif 1650 POISON_HEAP_REF w2 1651 str w2, [x3, #MIRROR_OBJECT_CLASS_OFFSET] 1652 // Fence. This is "ish" not "ishst" so 1653 // that it also ensures ordering of 1654 // the class status load with respect 1655 // to later accesses to the class 1656 // object. Alternatively we could use 1657 // "ishst" if we use load-acquire for 1658 // the class status load.) 1659 // Needs to be done before pushing on 1660 // allocation since Heap::VisitObjects 1661 // relies on seeing the class pointer. 1662 // b/28790624 1663 dmb ish 1664 // Push the new object onto the thread 1665 // local allocation stack and 1666 // increment the thread local 1667 // allocation stack top. 1668 ldr x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1669 str w3, [x1], #COMPRESSED_REFERENCE_SIZE // (Increment x1 as a side effect.) 1670 str x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1671 // Decrement the size of the free list 1672 ldr w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1673 sub x1, x1, #1 1674 // TODO: consider combining this store 1675 // and the list head store above using 1676 // strd. 1677 str w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1678 1679 mov x0, x3 // Set the return value and return. 1680 ret 1681.Lart_quick_alloc_object_rosalloc_slow_path: 1682 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1683 mov x2, xSELF // pass Thread::Current 1684 bl artAllocObjectFromCodeRosAlloc // (uint32_t type_idx, Method* method, Thread*) 1685 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1686 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1687END art_quick_alloc_object_rosalloc 1688 1689// The common fast path code for art_quick_alloc_object_tlab and art_quick_alloc_object_region_tlab. 1690// 1691// x0: type_idx/return value, x1: ArtMethod*, x2: Class*, xSELF(x19): Thread::Current 1692// x3-x7: free. 1693// Need to preserve x0 and x1 to the slow path. 1694.macro ALLOC_OBJECT_TLAB_FAST_PATH slowPathLabel 1695 cbz x2, \slowPathLabel // Check null class 1696 // Check class status. 1697 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] 1698 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 1699 bne \slowPathLabel 1700 // Add a fake dependence from the 1701 // following access flag and size 1702 // loads to the status load. 1703 // This is to prevent those loads 1704 // from being reordered above the 1705 // status load and reading wrong 1706 // values (an alternative is to use 1707 // a load-acquire for the status). 1708 eor x3, x3, x3 1709 add x2, x2, x3 1710 // Check access flags has 1711 // kAccClassIsFinalizable. 1712 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 1713 tbnz x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE_BIT, \slowPathLabel 1714 // Load thread_local_pos (x4) and 1715 // thread_local_end (x5). 1716 ldr x4, [xSELF, #THREAD_LOCAL_POS_OFFSET] 1717 ldr x5, [xSELF, #THREAD_LOCAL_END_OFFSET] 1718 sub x6, x5, x4 // Compute the remaining buf size. 1719 ldr w7, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x7). 1720 cmp x7, x6 // Check if it fits. OK to do this 1721 // before rounding up the object size 1722 // assuming the buf size alignment. 1723 bhi \slowPathLabel 1724 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1725 // Round up the object size by the 1726 // object alignment. (addr + 7) & ~7. 1727 add x7, x7, #OBJECT_ALIGNMENT_MASK 1728 and x7, x7, #OBJECT_ALIGNMENT_MASK_TOGGLED 1729 // Move old thread_local_pos to x0 1730 // for the return value. 1731 mov x0, x4 1732 add x5, x0, x7 1733 str x5, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 1734 ldr x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 1735 add x5, x5, #1 1736 str x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 1737 POISON_HEAP_REF w2 1738 str w2, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 1739 // Fence. This is "ish" not "ishst" so 1740 // that the code after this allocation 1741 // site will see the right values in 1742 // the fields of the class. 1743 // Alternatively we could use "ishst" 1744 // if we use load-acquire for the 1745 // class status load.) 1746 dmb ish 1747 ret 1748.endm 1749 1750// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_tlab, TLAB). 1751ENTRY art_quick_alloc_object_tlab 1752 // Fast path tlab allocation. 1753 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 1754 // x2-x7: free. 1755#if defined(USE_READ_BARRIER) 1756 mvn x0, xzr // Read barrier not supported here. 1757 ret // Return -1. 1758#endif 1759 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 1760 // Load the class (x2) 1761 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 1762 ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_tlab_slow_path 1763.Lart_quick_alloc_object_tlab_slow_path: 1764 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // Save callee saves in case of GC. 1765 mov x2, xSELF // Pass Thread::Current. 1766 bl artAllocObjectFromCodeTLAB // (uint32_t type_idx, Method* method, Thread*) 1767 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1768 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1769END art_quick_alloc_object_tlab 1770 1771// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_region_tlab, RegionTLAB) 1772ENTRY art_quick_alloc_object_region_tlab 1773 // Fast path region tlab allocation. 1774 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 1775 // x2-x7: free. 1776#if !defined(USE_READ_BARRIER) 1777 mvn x0, xzr // Read barrier must be enabled here. 1778 ret // Return -1. 1779#endif 1780 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 1781 // Load the class (x2) 1782 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 1783 // Read barrier for class load. 1784 ldr w3, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 1785 cbnz x3, .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path 1786.Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path_exit: 1787 ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_region_tlab_slow_path 1788.Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path: 1789 // The read barrier slow path. Mark 1790 // the class. 1791 stp x0, x1, [sp, #-32]! // Save registers (x0, x1, lr). 1792 str xLR, [sp, #16] // Align sp by 16 bytes. 1793 mov x0, x2 // Pass the class as the first param. 1794 bl artReadBarrierMark 1795 mov x2, x0 // Get the (marked) class back. 1796 ldp x0, x1, [sp, #0] // Restore registers. 1797 ldr xLR, [sp, #16] 1798 add sp, sp, #32 1799 b .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path_exit 1800.Lart_quick_alloc_object_region_tlab_slow_path: 1801 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // Save callee saves in case of GC. 1802 mov x2, xSELF // Pass Thread::Current. 1803 bl artAllocObjectFromCodeRegionTLAB // (uint32_t type_idx, Method* method, Thread*) 1804 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1805 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1806END art_quick_alloc_object_region_tlab 1807 1808 /* 1809 * Called by managed code when the thread has been asked to suspend. 1810 */ 1811 .extern artTestSuspendFromCode 1812ENTRY art_quick_test_suspend 1813 ldrh w0, [xSELF, #THREAD_FLAGS_OFFSET] // get xSELF->state_and_flags.as_struct.flags 1814 cbnz w0, .Lneed_suspend // check flags == 0 1815 ret // return if flags == 0 1816.Lneed_suspend: 1817 mov x0, xSELF 1818 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1819 bl artTestSuspendFromCode // (Thread*) 1820 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1821END art_quick_test_suspend 1822 1823ENTRY art_quick_implicit_suspend 1824 mov x0, xSELF 1825 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1826 bl artTestSuspendFromCode // (Thread*) 1827 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1828END art_quick_implicit_suspend 1829 1830 /* 1831 * Called by managed code that is attempting to call a method on a proxy class. On entry 1832 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 1833 * method agrees with a ref and args callee save frame. 1834 */ 1835 .extern artQuickProxyInvokeHandler 1836ENTRY art_quick_proxy_invoke_handler 1837 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 1838 mov x2, xSELF // pass Thread::Current 1839 mov x3, sp // pass SP 1840 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 1841 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 1842 cbnz x2, .Lexception_in_proxy // success if no exception is pending 1843 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Restore frame 1844 fmov d0, x0 // Store result in d0 in case it was float or double 1845 ret // return on success 1846.Lexception_in_proxy: 1847 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1848 DELIVER_PENDING_EXCEPTION 1849END art_quick_proxy_invoke_handler 1850 1851 /* 1852 * Called to resolve an imt conflict. 1853 * x0 is the conflict ArtMethod. 1854 * xIP1 is a hidden argument that holds the target interface method's dex method index. 1855 * 1856 * Note that this stub writes to xIP0, xIP1, and x0. 1857 */ 1858 .extern artInvokeInterfaceTrampoline 1859ENTRY art_quick_imt_conflict_trampoline 1860 ldr xIP0, [sp, #0] // Load referrer 1861 ldr xIP0, [xIP0, #ART_METHOD_DEX_CACHE_METHODS_OFFSET_64] // Load dex cache methods array 1862 ldr xIP0, [xIP0, xIP1, lsl #POINTER_SIZE_SHIFT] // Load interface method 1863 ldr xIP1, [x0, #ART_METHOD_JNI_OFFSET_64] // Load ImtConflictTable 1864 ldr x0, [xIP1] // Load first entry in ImtConflictTable. 1865.Limt_table_iterate: 1866 cmp x0, xIP0 1867 // Branch if found. Benchmarks have shown doing a branch here is better. 1868 beq .Limt_table_found 1869 // If the entry is null, the interface method is not in the ImtConflictTable. 1870 cbz x0, .Lconflict_trampoline 1871 // Iterate over the entries of the ImtConflictTable. 1872 ldr x0, [xIP1, #(2 * __SIZEOF_POINTER__)]! 1873 b .Limt_table_iterate 1874.Limt_table_found: 1875 // We successfully hit an entry in the table. Load the target method 1876 // and jump to it. 1877 ldr x0, [xIP1, #__SIZEOF_POINTER__] 1878 ldr xIP0, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 1879 br xIP0 1880.Lconflict_trampoline: 1881 // Call the runtime stub to populate the ImtConflictTable and jump to the 1882 // resolved method. 1883 INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline 1884END art_quick_imt_conflict_trampoline 1885 1886ENTRY art_quick_resolution_trampoline 1887 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1888 mov x2, xSELF 1889 mov x3, sp 1890 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 1891 cbz x0, 1f 1892 mov xIP0, x0 // Remember returned code pointer in xIP0. 1893 ldr x0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 1894 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1895 br xIP0 18961: 1897 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1898 DELIVER_PENDING_EXCEPTION 1899END art_quick_resolution_trampoline 1900 1901/* 1902 * Generic JNI frame layout: 1903 * 1904 * #-------------------# 1905 * | | 1906 * | caller method... | 1907 * #-------------------# <--- SP on entry 1908 * | Return X30/LR | 1909 * | X29/FP | callee save 1910 * | X28 | callee save 1911 * | X27 | callee save 1912 * | X26 | callee save 1913 * | X25 | callee save 1914 * | X24 | callee save 1915 * | X23 | callee save 1916 * | X22 | callee save 1917 * | X21 | callee save 1918 * | X20 | callee save 1919 * | X19 | callee save 1920 * | X7 | arg7 1921 * | X6 | arg6 1922 * | X5 | arg5 1923 * | X4 | arg4 1924 * | X3 | arg3 1925 * | X2 | arg2 1926 * | X1 | arg1 1927 * | D7 | float arg 8 1928 * | D6 | float arg 7 1929 * | D5 | float arg 6 1930 * | D4 | float arg 5 1931 * | D3 | float arg 4 1932 * | D2 | float arg 3 1933 * | D1 | float arg 2 1934 * | D0 | float arg 1 1935 * | Method* | <- X0 1936 * #-------------------# 1937 * | local ref cookie | // 4B 1938 * | handle scope size | // 4B 1939 * #-------------------# 1940 * | JNI Call Stack | 1941 * #-------------------# <--- SP on native call 1942 * | | 1943 * | Stack for Regs | The trampoline assembly will pop these values 1944 * | | into registers for native call 1945 * #-------------------# 1946 * | Native code ptr | 1947 * #-------------------# 1948 * | Free scratch | 1949 * #-------------------# 1950 * | Ptr to (1) | <--- SP 1951 * #-------------------# 1952 */ 1953 /* 1954 * Called to do a generic JNI down-call 1955 */ 1956ENTRY art_quick_generic_jni_trampoline 1957 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 1958 1959 // Save SP , so we can have static CFI info. 1960 mov x28, sp 1961 .cfi_def_cfa_register x28 1962 1963 // This looks the same, but is different: this will be updated to point to the bottom 1964 // of the frame when the handle scope is inserted. 1965 mov xFP, sp 1966 1967 mov xIP0, #5120 1968 sub sp, sp, xIP0 1969 1970 // prepare for artQuickGenericJniTrampoline call 1971 // (Thread*, SP) 1972 // x0 x1 <= C calling convention 1973 // xSELF xFP <= where they are 1974 1975 mov x0, xSELF // Thread* 1976 mov x1, xFP 1977 bl artQuickGenericJniTrampoline // (Thread*, sp) 1978 1979 // The C call will have registered the complete save-frame on success. 1980 // The result of the call is: 1981 // x0: pointer to native code, 0 on error. 1982 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 1983 1984 // Check for error = 0. 1985 cbz x0, .Lexception_in_native 1986 1987 // Release part of the alloca. 1988 mov sp, x1 1989 1990 // Save the code pointer 1991 mov xIP0, x0 1992 1993 // Load parameters from frame into registers. 1994 // TODO Check with artQuickGenericJniTrampoline. 1995 // Also, check again APPCS64 - the stack arguments are interleaved. 1996 ldp x0, x1, [sp] 1997 ldp x2, x3, [sp, #16] 1998 ldp x4, x5, [sp, #32] 1999 ldp x6, x7, [sp, #48] 2000 2001 ldp d0, d1, [sp, #64] 2002 ldp d2, d3, [sp, #80] 2003 ldp d4, d5, [sp, #96] 2004 ldp d6, d7, [sp, #112] 2005 2006 add sp, sp, #128 2007 2008 blr xIP0 // native call. 2009 2010 // result sign extension is handled in C code 2011 // prepare for artQuickGenericJniEndTrampoline call 2012 // (Thread*, result, result_f) 2013 // x0 x1 x2 <= C calling convention 2014 mov x1, x0 // Result (from saved). 2015 mov x0, xSELF // Thread register. 2016 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 2017 2018 bl artQuickGenericJniEndTrampoline 2019 2020 // Pending exceptions possible. 2021 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2022 cbnz x2, .Lexception_in_native 2023 2024 // Tear down the alloca. 2025 mov sp, x28 2026 .cfi_def_cfa_register sp 2027 2028 // Tear down the callee-save frame. 2029 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2030 2031 // store into fpr, for when it's a fpr return... 2032 fmov d0, x0 2033 ret 2034 2035.Lexception_in_native: 2036 // Move to x1 then sp to please assembler. 2037 ldr x1, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 2038 mov sp, x1 2039 .cfi_def_cfa_register sp 2040 # This will create a new save-all frame, required by the runtime. 2041 DELIVER_PENDING_EXCEPTION 2042END art_quick_generic_jni_trampoline 2043 2044/* 2045 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 2046 * of a quick call: 2047 * x0 = method being called/to bridge to. 2048 * x1..x7, d0..d7 = arguments to that method. 2049 */ 2050ENTRY art_quick_to_interpreter_bridge 2051 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Set up frame and save arguments. 2052 2053 // x0 will contain mirror::ArtMethod* method. 2054 mov x1, xSELF // How to get Thread::Current() ??? 2055 mov x2, sp 2056 2057 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 2058 // mirror::ArtMethod** sp) 2059 bl artQuickToInterpreterBridge 2060 2061 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // TODO: no need to restore arguments in this case. 2062 2063 fmov d0, x0 2064 2065 RETURN_OR_DELIVER_PENDING_EXCEPTION 2066END art_quick_to_interpreter_bridge 2067 2068 2069// 2070// Instrumentation-related stubs 2071// 2072 .extern artInstrumentationMethodEntryFromCode 2073ENTRY art_quick_instrumentation_entry 2074 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2075 2076 mov x20, x0 // Preserve method reference in a callee-save. 2077 2078 mov x2, xSELF 2079 mov x3, xLR 2080 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, LR) 2081 2082 mov xIP0, x0 // x0 = result of call. 2083 mov x0, x20 // Reload method reference. 2084 2085 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Note: will restore xSELF 2086 adr xLR, art_quick_instrumentation_exit 2087 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 2088END art_quick_instrumentation_entry 2089 2090 .extern artInstrumentationMethodExitFromCode 2091ENTRY art_quick_instrumentation_exit 2092 mov xLR, #0 // Clobber LR for later checks. 2093 2094 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 2095 2096 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 2097 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 2098 // easier to have an extra small stack area. 2099 2100 str x0, [sp, #-16]! // Save integer result. 2101 .cfi_adjust_cfa_offset 16 2102 str d0, [sp, #8] // Save floating-point result. 2103 2104 add x1, sp, #16 // Pass SP. 2105 mov x2, x0 // Pass integer result. 2106 fmov x3, d0 // Pass floating-point result. 2107 mov x0, xSELF // Pass Thread. 2108 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 2109 2110 mov xIP0, x0 // Return address from instrumentation call. 2111 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 2112 2113 ldr d0, [sp, #8] // Restore floating-point result. 2114 ldr x0, [sp], 16 // Restore integer result, and drop stack area. 2115 .cfi_adjust_cfa_offset 16 2116 2117 POP_REFS_ONLY_CALLEE_SAVE_FRAME 2118 2119 br xIP0 // Tail-call out. 2120END art_quick_instrumentation_exit 2121 2122 /* 2123 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 2124 * will long jump to the upcall with a special exception of -1. 2125 */ 2126 .extern artDeoptimize 2127ENTRY art_quick_deoptimize 2128 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 2129 mov x0, xSELF // Pass thread. 2130 bl artDeoptimize // artDeoptimize(Thread*) 2131 brk 0 2132END art_quick_deoptimize 2133 2134 /* 2135 * Compiled code has requested that we deoptimize into the interpreter. The deoptimization 2136 * will long jump to the upcall with a special exception of -1. 2137 */ 2138 .extern artDeoptimizeFromCompiledCode 2139ENTRY art_quick_deoptimize_from_compiled_code 2140 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 2141 mov x0, xSELF // Pass thread. 2142 bl artDeoptimizeFromCompiledCode // artDeoptimizeFromCompiledCode(Thread*) 2143 brk 0 2144END art_quick_deoptimize_from_compiled_code 2145 2146 2147 /* 2148 * String's indexOf. 2149 * 2150 * TODO: Not very optimized. 2151 * On entry: 2152 * x0: string object (known non-null) 2153 * w1: char to match (known <= 0xFFFF) 2154 * w2: Starting offset in string data 2155 */ 2156ENTRY art_quick_indexof 2157 ldr w3, [x0, #MIRROR_STRING_COUNT_OFFSET] 2158 add x0, x0, #MIRROR_STRING_VALUE_OFFSET 2159 2160 /* Clamp start to [0..count] */ 2161 cmp w2, #0 2162 csel w2, wzr, w2, lt 2163 cmp w2, w3 2164 csel w2, w3, w2, gt 2165 2166 /* Save a copy to compute result */ 2167 mov x5, x0 2168 2169 /* Build pointer to start of data to compare and pre-bias */ 2170 add x0, x0, x2, lsl #1 2171 sub x0, x0, #2 2172 2173 /* Compute iteration count */ 2174 sub w2, w3, w2 2175 2176 /* 2177 * At this point we have: 2178 * x0: start of the data to test 2179 * w1: char to compare 2180 * w2: iteration count 2181 * x5: original start of string data 2182 */ 2183 2184 subs w2, w2, #4 2185 b.lt .Lindexof_remainder 2186 2187.Lindexof_loop4: 2188 ldrh w6, [x0, #2]! 2189 ldrh w7, [x0, #2]! 2190 ldrh wIP0, [x0, #2]! 2191 ldrh wIP1, [x0, #2]! 2192 cmp w6, w1 2193 b.eq .Lmatch_0 2194 cmp w7, w1 2195 b.eq .Lmatch_1 2196 cmp wIP0, w1 2197 b.eq .Lmatch_2 2198 cmp wIP1, w1 2199 b.eq .Lmatch_3 2200 subs w2, w2, #4 2201 b.ge .Lindexof_loop4 2202 2203.Lindexof_remainder: 2204 adds w2, w2, #4 2205 b.eq .Lindexof_nomatch 2206 2207.Lindexof_loop1: 2208 ldrh w6, [x0, #2]! 2209 cmp w6, w1 2210 b.eq .Lmatch_3 2211 subs w2, w2, #1 2212 b.ne .Lindexof_loop1 2213 2214.Lindexof_nomatch: 2215 mov x0, #-1 2216 ret 2217 2218.Lmatch_0: 2219 sub x0, x0, #6 2220 sub x0, x0, x5 2221 asr x0, x0, #1 2222 ret 2223.Lmatch_1: 2224 sub x0, x0, #4 2225 sub x0, x0, x5 2226 asr x0, x0, #1 2227 ret 2228.Lmatch_2: 2229 sub x0, x0, #2 2230 sub x0, x0, x5 2231 asr x0, x0, #1 2232 ret 2233.Lmatch_3: 2234 sub x0, x0, x5 2235 asr x0, x0, #1 2236 ret 2237END art_quick_indexof 2238 2239 /* 2240 * Create a function `name` calling the ReadBarrier::Mark routine, 2241 * getting its argument and returning its result through W register 2242 * `wreg` (corresponding to X register `xreg`), saving and restoring 2243 * all caller-save registers. 2244 * 2245 * If `wreg` is different from `w0`, the generated function follows a 2246 * non-standard runtime calling convention: 2247 * - register `wreg` is used to pass the (sole) argument of this 2248 * function (instead of W0); 2249 * - register `wreg` is used to return the result of this function 2250 * (instead of W0); 2251 * - W0 is treated like a normal (non-argument) caller-save register; 2252 * - everything else is the same as in the standard runtime calling 2253 * convention (e.g. standard callee-save registers are preserved). 2254 */ 2255.macro READ_BARRIER_MARK_REG name, wreg, xreg 2256ENTRY \name 2257 /* 2258 * Allocate 46 stack slots * 8 = 368 bytes: 2259 * - 20 slots for core registers X0-X19 2260 * - 24 slots for floating-point registers D0-D7 and D16-D31 2261 * - 1 slot for return address register XLR 2262 * - 1 padding slot for 16-byte stack alignment 2263 */ 2264 // Save all potentially live caller-save core registers. 2265 stp x0, x1, [sp, #-368]! 2266 .cfi_adjust_cfa_offset 368 2267 .cfi_rel_offset x0, 0 2268 .cfi_rel_offset x1, 8 2269 stp x2, x3, [sp, #16] 2270 .cfi_rel_offset x2, 16 2271 .cfi_rel_offset x3, 24 2272 stp x4, x5, [sp, #32] 2273 .cfi_rel_offset x4, 32 2274 .cfi_rel_offset x5, 40 2275 stp x6, x7, [sp, #48] 2276 .cfi_rel_offset x6, 48 2277 .cfi_rel_offset x7, 56 2278 stp x8, x9, [sp, #64] 2279 .cfi_rel_offset x8, 64 2280 .cfi_rel_offset x9, 72 2281 stp x10, x11, [sp, #80] 2282 .cfi_rel_offset x10, 80 2283 .cfi_rel_offset x11, 88 2284 stp x12, x13, [sp, #96] 2285 .cfi_rel_offset x12, 96 2286 .cfi_rel_offset x13, 104 2287 stp x14, x15, [sp, #112] 2288 .cfi_rel_offset x14, 112 2289 .cfi_rel_offset x15, 120 2290 stp x16, x17, [sp, #128] 2291 .cfi_rel_offset x16, 128 2292 .cfi_rel_offset x17, 136 2293 stp x18, x19, [sp, #144] 2294 .cfi_rel_offset x18, 144 2295 .cfi_rel_offset x19, 152 2296 // Save all potentially live caller-save floating-point registers. 2297 stp d0, d1, [sp, #160] 2298 stp d2, d3, [sp, #176] 2299 stp d4, d5, [sp, #192] 2300 stp d6, d7, [sp, #208] 2301 stp d16, d17, [sp, #224] 2302 stp d18, d19, [sp, #240] 2303 stp d20, d21, [sp, #256] 2304 stp d22, d23, [sp, #272] 2305 stp d24, d25, [sp, #288] 2306 stp d26, d27, [sp, #304] 2307 stp d28, d29, [sp, #320] 2308 stp d30, d31, [sp, #336] 2309 // Save return address. 2310 str xLR, [sp, #352] 2311 .cfi_rel_offset x30, 352 2312 // (sp + #360 is a padding slot) 2313 2314 .ifnc \wreg, w0 2315 mov w0, \wreg // Pass arg1 - obj from `wreg` 2316 .endif 2317 bl artReadBarrierMark // artReadBarrierMark(obj) 2318 .ifnc \wreg, w0 2319 mov \wreg, w0 // Return result into `wreg` 2320 .endif 2321 2322 // Restore core regs, except `xreg`, as `wreg` is used to return the 2323 // result of this function (simply remove it from the stack instead). 2324 POP_REGS_NE x0, x1, 0, \xreg 2325 POP_REGS_NE x2, x3, 16, \xreg 2326 POP_REGS_NE x4, x5, 32, \xreg 2327 POP_REGS_NE x6, x7, 48, \xreg 2328 POP_REGS_NE x8, x9, 64, \xreg 2329 POP_REGS_NE x10, x11, 80, \xreg 2330 POP_REGS_NE x12, x13, 96, \xreg 2331 POP_REGS_NE x14, x15, 112, \xreg 2332 POP_REGS_NE x16, x17, 128, \xreg 2333 POP_REGS_NE x18, x19, 144, \xreg 2334 // Restore floating-point registers. 2335 ldp d0, d1, [sp, #160] 2336 ldp d2, d3, [sp, #176] 2337 ldp d4, d5, [sp, #192] 2338 ldp d6, d7, [sp, #208] 2339 ldp d16, d17, [sp, #224] 2340 ldp d18, d19, [sp, #240] 2341 ldp d20, d21, [sp, #256] 2342 ldp d22, d23, [sp, #272] 2343 ldp d24, d25, [sp, #288] 2344 ldp d26, d27, [sp, #304] 2345 ldp d28, d29, [sp, #320] 2346 ldp d30, d31, [sp, #336] 2347 // Restore return address and remove padding. 2348 ldr xLR, [sp, #352] 2349 .cfi_restore x30 2350 add sp, sp, #368 2351 .cfi_adjust_cfa_offset -368 2352 ret 2353END \name 2354.endm 2355 2356READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg00, w0, x0 2357READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg01, w1, x1 2358READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg02, w2, x2 2359READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg03, w3, x3 2360READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg04, w4, x4 2361READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg05, w5, x5 2362READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg06, w6, x6 2363READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg07, w7, x7 2364READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg08, w8, x8 2365READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg09, w9, x9 2366READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg10, w10, x10 2367READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg11, w11, x11 2368READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg12, w12, x12 2369READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg13, w13, x13 2370READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg14, w14, x14 2371READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg15, w15, x15 2372READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg16, w16, x16 2373READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg17, w17, x17 2374READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg18, w18, x18 2375READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg19, w19, x19 2376READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg20, w20, x20 2377READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg21, w21, x21 2378READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg22, w22, x22 2379READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg23, w23, x23 2380READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg24, w24, x24 2381READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg25, w25, x25 2382READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg26, w26, x26 2383READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg27, w27, x27 2384READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg28, w28, x28 2385READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg29, w29, x29 2386